Integrating Algorithms into Neural Networks

Classic algorithms and machine learning systems like neural networks

are both abundant in everyday life. While classic computer

science algorithms are suitable for precise execution of exactly defined

tasks such as finding the shortest path in a large graph, neural

networks allow learning from data to predict the most likely answer in

more complex tasks such as image classification, which cannot be

reduced to an exact algorithm. In the talk, we explore combining

both concepts leading to more robust, better performing, more interpretable,

more computationally efficient, and most importantly data

efficient architectures. Using algorithmic supervision a neural network can learn

from or in conjunction with an algorithm. When integrating an algorithm

into a neural architecture, it is important that the algorithm is

differentiable such that the architecture can be trained end-to-end and

gradients can be propagated back through the algorithm in a meaningful

way. To make algorithms differentiable, I discuss a general

method for continuously relaxing algorithms by perturbing variables

with logistic distributions. In addition, I discuss specialized differentiable

algorithms such as differentiable sorting networks, and efficient

and effective differentiable sorting and ranking operators allowing

sorting and ranking supervision. Furthermore, I delve into differentiable

rendering, specifically, the generalized differentiable renderer GenDR.